Goals decide hockey games and mislead hockey bettors. Scoring is rare enough that a couple of weeks of results says almost nothing about how well a team is actually playing, which is why the numbers that predict NHL outcomes mostly describe what happens before the puck crosses the line. This piece covers the short list that earns attention; for how they feed into actual markets and prices, the NHL betting guide covers the other half.
Why do shot attempts predict better than goals?
Sample size, mostly. A game produces around six goals but well over a hundred shot attempts, so attempt-based numbers stabilise in weeks while goal-based numbers take months. Territorial control — who spends the night in whose end — repeats from game to game. Finishing does not: shooting percentages run hot and cold in streaks long enough to fool anyone watching only the scoreboard.
Shot-attempt share, often labelled Corsi, is the crude version: every attempt counted equally. Expected goals, or xG, is the refined one — each attempt weighted by how often that kind of chance goes in, based on location, shot type and context. A team's expected-goals share is the best single public number for how good it is right now, and the gap between a team's actual goals and its expected goals is a decent first guess at what is about to change.
Neither is magic. They are built from public data, every book and serious bettor uses them, and they describe team play rather than tonight's lineup or goalie. Their real value is as a discipline: they stop you believing the last five results.
What is PDO and why is it called the luck meter?
PDO is team shooting percentage plus team save percentage, usually taken at even strength. Across the whole league it has to average roughly 100 by construction, because every shot on goal is either a goal for one side or a save for the other. A team sitting well above 100 is getting elevated finishing, elevated goaltending, or both. Well below, the reverse.
The label is shorthand rather than literal truth. Elite shooters and an elite goalie can hold a team a little above the middle for long stretches. But large deviations are blunt about their future: they come back toward 100 far more often than they persist.
The betting use is specific. PDO flags teams whose results have detached from their play. A side winning games on a sky-high PDO with a mediocre expected-goals share is exactly the kind of team the standings oversell and the market sometimes overprices. The reverse profile — strong underlying play, ugly recent results — is where value tends to hide. PDO does not tell you what happens next; it tells you which teams' records to distrust.
How stable are special teams and goaltending numbers?
Power-play and penalty-kill percentages are loud, widely quoted, and unstable over short stretches. Conversion rates bounce; what persists better is the underlying machinery — how many chances a power play generates, how well a kill suppresses them. A power play creating plenty and converting nothing is usually a better bet going forward than one riding a hot conversion streak, even though the headline percentages say the opposite.
Goaltending is the extreme case: the biggest single variable in any hockey outcome, attached to the least reliable stat. Save percentage needs seasons rather than weeks to mean much — the full argument is in NHL starting goalies and betting. A rough ordering of trustworthiness over a month of hockey, from most to least:
- Expected-goals share and shot-attempt share
- Power-play chance creation and penalty-kill suppression
- Special-teams conversion percentages
- Goaltending and shooting percentages
What are score effects and why do they distort shot counts?
Teams change behaviour with the scoreline. Leading teams sit back, protect the middle of the ice and concede shots from the outside; trailing teams push, take risks and pile up attempts. So a team that trailed all night finishes with gaudy shot totals earned against an opponent that had stopped attacking. Raw counts flatter losers and undersell winners.
This is why serious versions of these stats are score- and venue-adjusted, and why a box score reading 38 shots to 22 means little without the game state attached. Reading one game honestly means asking who led, from when, and what the shot flow looked like while the score was still close. Score effects are also part of why late-game markets behave as they do, with trailing teams generating chances right up until the goalie comes off — mechanics that settle a lot of puck line bets in the final two minutes.
| Stat | What it measures | How fast it stabilises | Best use |
|---|---|---|---|
| Goals / results | what happened | very slowly | almost nothing predictive |
| Shot-attempt share | territorial control | quickly | baseline team strength |
| Expected goals (xG) | volume plus chance quality | quickly | best single team number |
| Shooting % | finishing | very slowly | spotting streaks likely to fade |
| PDO | shooting % + save % | a flag, not a skill | distrust-check on the standings |
None of these numbers picks winners on its own — the market sees them too. What they do is keep you honest: about which teams are actually good, which records are inflated, and which of your opinions are just the last two weeks of bounces talking. Where those opinions meet real prices is the subject of the main NHL betting guide.